چكيده به لاتين
Nowadays, everybody explicitly confirms the significance of analyzing social networks in order to understand social systems, and to help decision making. Networks are very complex, and multi-dimensional. In order to analyze such networks, especially in case of online social networks, detecting and studying the underlying communities might be the primary step. These patterns, i.e., communities, are not only caused by users’ interactions, but also influenced by other dimensions, such as users’ attributes, contents, or information flows.
This dissertation, in order to develop a new model for detecting communities in multi-dimensional networks, at first, reviewed the literature, and recognized gaps. Recognizing gaps could make opportunity for studying one-dimension networks and reviewing the present approaches of community detection. Then, in order to do a research on bi-dimensional and multi-dimensional networks, effective approaches would be selected and used. On the other hand, since evaluation (and verification) of methods emerges as a challenging issue, some well-known evaluation methods would be analyzed.
Community detection models and algorithms developed for multi-dimensional networks can only work on bi-dimensional ones; however, real social networks are multi-dimensional. This dissertation proposed a multi-objective model for coping with this issue. The main achievement of this model is that users’ behavior might be crucially dissimilar through various dimensions. This dissimilarity tells us about the significance of considering different dimensions in our analysis. Moreover, studying such finding could be as an innovation in “concept” layer, which needs new modeling and new algorithms. Based on this requirement, the dissertation proposed a consensus-based community detection approach (CBC). According to CBC, communities present in each dimension should be detected, then, the final communities could be extracted from them by means of a consensus-based clustering method. This approach provides a line of advantages, such as the capability of simultaneously analyzing of a set of dimensions, calculating effect of each dimension on the final results, and handling missing values.